Research And Application Of Recommendation Based On Item Classifying And User’s Context

With the fast development of Internet and prosperity of e-commerce, peopleusually spend a long time on looking for their really needed resource when they facewith lots of resources, which brings inconvenience to today fast packed people. Theemergence of the recommendation technology well solves this problem and it attractsmore and more attention from people. Recommendation system initiativelyrecommends resource to users by users’ personal information, history behavior and soon, which helps users save time when they select resources.AS an efficient way to solve the problem of too many resources, recommendationtechnology has developed many kinds nowadays. These ones use much knowledge fromother areas, such as Data Mining, Neural Network, Artificial Intelligence and so on.Usually, Recommendation technology contains based on association rules, based oncontent, based on collaborative filter, based on mix and other personalizedrecommendation system. Every personalized recommendation technology has itsdisadvantage and advantage. It will produce different result on different object.Based on collaborative filter recommendation technology is one of them. It is usedthe most widely and successfully. It uses items’ rating to look for the nearest neighborsfor target people or item and makes use of the nearest neighbors’ ratings to computepredicted rating for the target item. The system decides whether to recommend thetarget item to user according to the predicted rating. However, based on collaborativefilter personalized recommendation technology has disadvantages such as datasparseness, cold starting, which affect the recommendation quality.This paper has two main points. The first point is a method which advancescomputing of item similarity by using feature attributes and ratings. The second point isa recommendation technology which is based on item classifying and user’s context,which joins item classify and user’s context into recommendation technology based oncollaborative filter. About the first point, it comprehensives item’s feature attributesfrom recommendation technology based on content and item’s rating fromrecommendation technology based on collaborative filter. Firstly, it uses the featureattributes of items to classify items. Secondly, it looks for the nearest neighbors of targetitem just in item category which contains target item. This method is different from thetraditional collaborative filter personalized recommendation technology which selects the nearest neighbors of target item from all items. It sufficient takes advantage offeature attributes and rating of item, which advances the reliability of the nearestneighbors. About the second point, it introduce users’ personalized context and buildsthe corresponding relation of item category and users’ context. When system predictsthe rating of the target item for the target user, it finds the item category which the targetitem belongs to according to the feature attributes and the personalized context of thetarget user for this item category. Then, system uses the ratings of users who has thesame context with the target user and the nearest neighbors of the target item to computepredicted rating of the target item.This text uses MovieLens dataset to test the modified method it presents and takethe experimental result to compare with Slope One. The result indicates that the methodof this text is better than Slope One in accuracy of recommendation. Last, the method ofthis text is used in a practical project.